Black Box Software Testing Spring 2005 MULTIVARIABLE TESTING
Black Box Software Testing Spring 2005 MULTI-VARIABLE TESTING by Cem Kaner, J. D. , Ph. D. Professor of Software Engineering Florida Institute of Technology and James Bach Principal, Satisfice Inc. Copyright (c) Cem Kaner & James Bach, 2000 -2005 This work is licensed under the Creative Commons Attribution-Share. Alike License. To view a copy of this license, visit http: //creativecommons. org/licenses/by-sa/2. 0/ or send a letter to Creative Commons, 559 Nathan Abbott Way, Stanford, California 94305, USA. These notes are partially based on research that was supported by NSF Grant EIA-0113539 ITR/SY+PE: "Improving the Education of Software Testers. " Any opinions, findings and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation. Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 1
Combination Chart Var 1 Var 2 Var 3 Var 4 Var 5 Test 1 Value 12 Value 13 Value 14 Value 15 Test 2 Value 21 Value 22 Value 23 Value 24 Value 25 Test 3 Value 31 Value 32 Value 33 Value 34 Value 35 Test 4 Value 41 Value 42 Value 43 Value 44 Value 45 Test 5 Value 51 Value 52 Value 53 Value 54 Value 55 Test 6 Value 61 Value 62 Value 63 Value 64 Value 65 In a combination test, we test several variables together. Each test explicitly sets values for each of the variables under test. Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 2
Challenges of multivariable testing 1. 2. 3. 4. The space of possible tests is enormous How to decide which variables to combine? What IS the fault model? How to figure out what the relationships among the variables actually are, in detail? Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 3
Combination Testing • There are several approaches to combination testing: – Mechanical (or procedural). The tester uses a routine procedure to determine a good set of tests – Risk-based. The tester combines test values (the values of each variable) based on perceived risks associated with noteworthy combinations – Scenario-based. The tester combines test values on the basis of interesting stories created for the combinations Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 4
Domain testing • In 1 -dimensional testing, we run two tests for every boundary: – Test the boundary-valid value – Test the boundary-invalid value • For example if X < 24 defines the boundary, we use X=24 (invalid) and X=24 -delta (valid). We choose the smallest workable delta to minimize the possibility of an error hiding within the gap between 24 and 24 -delta. • In multi-dimensional testing, we start by testing each dimension on its own, reasonably thoroughly. • Then we reduce the set of values to test per dimension, probably to the boundaries: – Too-low (TL), valid lowest (VL), valid biggest (VB), too big (TB) – where TL = VL-delta and TB=VB+delta Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 5
Defining the domain: 2 variables • Suppose we have two numeric variables, V 1 and V 2. • We analyze each variable in terms of its subdomains and boundaries. Thus we have for each variable: – V 1: Too-low (TL), valid lowest (VL), valid biggest (VB), too big (TB) – V 2: Too-low (TL), valid lowest (VL), valid biggest (VB), too big (TB) – Where we set • TL = VL-delta (smallest available difference between two numbers) • TH = VB+delta Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 6
Example: 2 variables Consider the following domain definition: 1 <= V 1 < 4 1 <= V 2 < 4 Store data to 3 digits precision: TL = 0. 999 VL = 1. 00 VB = 3. 99 TB = 4. 00 Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 7
Defining the domain 3 independent variables • Suppose we have 3 numeric variables, V 1, V 2, V 3. • We analyze each variable in terms of its subdomains and boundaries. Thus we might have for each variable: – V 1: Too-low (TL), valid lowest (VL), valid biggest (VB), too big (TB) – V 2: Too-low (TL), valid lowest (VL), valid biggest (VB), too big (TB) – V 3: Too-low (TL), valid lowest (VL), valid biggest (VB), too big (TB) In this simple model, anything inside the box is a valid value, and anything outside the box is not. (When we restrict ourselves to valid values, we are thinking inside the box. ) Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 8
Mechanical approach #1 "Weak testing“ version 1 V 2 V 3 Test 1 VL VL VL Test 2 VB VB VB Test 3 TL TL TL Test 4 TB TB TB • We create enough tests to cover every value of every variable, once. If the largest number of values is N, we need only N tests • Note the collisions of error cases. If Test 3 fails, is it because of the bad value of V 1, V 2, V 3, or some combination of them? • What bug do we expect to find in Test 3 that we would not find in a test of single dimension, with a bad value? Why do we need a combination? Too-low (TL), lowest valid (VL), biggest valid (VB), too big (TB) Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 9
"Weak testing" version 2 V 1 V 2 V 3 Test 1 VL VL VL Test 2 VB VB VB Test 3 TL VL VB Test 4 VB TL VL Test 5 VL VL TL Test 6 TB VB VL Test 7 VL TB VB Test 8 VB VL TB • In this second version, we treat error cases specially: – Generate a core set of tests for "valid" (non-error) inputs – Generate additional tests in which one error case is allowed per test case. (Jorgensen calls this “weak robust equivalence class testing. ” – We might also add a few market-critical combinations Too-low (TL), lowest valid (VL), biggest valid (VB), too big (TB) Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 10
"Weak testing" version 3 - All Singles V 1 V 2 V 3 Test 1 LV LV LV Test 2 BV BV BV • Drop the error cases – test them in single-variable tests. • Create tests only for valid values – Jorgensen calls this “weak normal equivalence class testing” • Note the coverage that we do and do not achieve: – We have a test for every valid value of interest of every variable – We are not set up to detect interactions among variables. • Here, for example, we check all minima together and all maxima. • Should we worry about Low-High combinations? Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 11
Mechanical approach #2 "Strong testing" version 1 Test every combination of values of interest: Jorgensen calls this "strong robust equivalence class testing" V 1 V 2 V 3 Test 1 TL TL TL Test 2 TL TL LV Test 3 TL TL BV Test 4 TL TL TB Test 5 TL LV TL Test 6 TL LV LV Test 7 TL LV BV Test 8 TL LV TB Test 9 TL BV TL Test 10 TL BV LV This is part of the table. The complete table has 4 * 4 tests. In general, if N is the number of variables we test together and they have k 1, k 2 … k. N values, strong testing requires k 1 x k 2 x … x k. N tests Too-low (TL), lowest valid (VL), biggest valid (VB), too big (TB) Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 12
"Strong testing" version 2 All n-tuples • Start with strong testing • But restrict the values of interest to valid values. – Jorgensen calls this “strong normal equivalence class analysis” • Cover error cases in the onevariable tests. • If there are N independent dimensions, and we test only LV and BV for each, there are 2 N tests Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 13
More “strong testing" • Another variation includes all valid-value combinations plus a separate set of combination tests in which one, some, or all variables have an error value. • Tests that include several errors are of interest only if we think that multiple errors might have some type of cumulative effect. Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 14
Mechanical approach #3 Combinatorial testing • We have N variables • We assume the variables are independent – A value of one variable does not change the effects or validity of values of other variables • We consider only valid values of interest – An invalid value stops the test. (V 1, V 2, Bad, V 4, V 5) what do we learn about V 1, V 2, V 4 or V 5? Anything in this test of interest other than “Bad” will be masked • Our goal is to sample from the space of possible N-tuples in way that assures a minimum level of combination coverage: – All N-tuples all combinations of valid values – All singles all individual valid values – All pairs all pairs of valid values – All triples all triplets of valid values Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 15
Combinatorial Example • Here is a simple Find dialog. It takes three inputs: – Find what: a text string – Match case: yes or no – Direction: up or down • Simplify this by considering only three values for the text string, “lowercase” and “Mixed Cases” and “CAPITALS”. Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 16
Combinations Example 1 How many combinations of these three variables are possible? 2 List ALL the combinations of these three variables. 3 Now create combination tests that cover all possible pairs of values, but don’t try to cover all possible triplets. List one such set. 4 How many test cases are in this set? Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 17
Combinations Example 1. How many combinations of these three variables are possible? – Find what has 3 values (lowercase, mixed, caps) (L M C) – Match case has 2 values (Yes / No) (Y N) – Direction has 2 values (Up / Down) (U D) So there will by 3 x 2 = 12 tests 2. List ALL the combinations of these three variables. LYU MYU CYU LYD MYD CYD LNU MNU CNU LND MND CND 3. By the way, a more complete analysis will also consider whether the string is in the document or not. We’ll add a 4 th binary variable to the analysis soon. Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 18
Building an all-pairs table • Label the columns with the variable names. • List variables in descending order (of number of possible values) • Each column will have repetition. – To determine how many times (rows in which) to repeat the first value before creating a row for the second multiply the number of variable values in column 1 x the number that will be in column 2 • In our example, – Find What has 3 values – Match Case has 2 values – So there will be at least 6 rows Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 19
Combination Testing • Building an all-pairs combination table: – In the second column, list all the values of the second variable, skip the line, list the values again, etc. In our example, variable 2’s possible values are U, D so the table looks like this so far Find (LMC) Match (YN) L Y L N M Y M N C Y C N Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 20
Combination Testing Building an all-pairs combination table: – Each section of the third column (think of LL as defining a section, MM as defining another) will have to contain every value of variable 3. Order the values such that the variables also make all pairs with variable 2. – Our variable 3 has two values, U and D – The third section can be filled in either way, and you might highlight it so that you can reverse it later. The decision (say D, U) is arbitrary. Black Box Software Testing Copyright © 2005 Find (LMC) Match (YN) Direct (UD) L Y U L N D M Y D M N U C Y D C N U Cem Kaner & James Bach 21
Combination Testing Now that we’ve solved the 3 -column exercise, let’s try adding more variables. Each will have two values. Let’s start by making this look a little more general A D F A E G B D G B E F C D G C E F Black Box Software Testing Copyright © 2005 The 4 th column goes in easily: • We start by making sure we hit all pairs of values of column 4 and column 2 • then all pairs of column 4 and column 3. A D F H A E G I B D G I B E F H C D G H C E F I Cem Kaner & James Bach 22
Combination Testing Watch this first attempt on column 5. We achieve all pairs of JK with columns 1, 2, and 3, but miss it for column 4. The most recent arbitrary choice was KJ in the 2 nd section. (Once that was determined, we had to pick JK for the third in order to pair K with an F in the 3 rd column. ) So we will erase the last choice and try again: Black Box Software Testing Copyright © 2005 A D F H J A E G I K B D G I K B E F H J C D G H J C E F I K Cem Kaner & James Bach 23
Combination Testing • We flipped the last arbitrary choice (column 5, section 2, to JK from KJ) and erased the JK in section 3. • We then fill in section 3 by checking for missing pairs. • JK, JK gives us three DJ, DJ pairs (2 nd and 5 th columns) so we have to flip to KJ for the third section. • Now everything works Black Box Software Testing Copyright © 2005 A D F H J A E G I K B D G I J B E F H K C D G H K C E F I J Cem Kaner & James Bach 24
Combination Testing But when we add the next column, we see that we just can’t achieve all pairs with 6 values. The first one works up to column 4 but then fails to get pair KL or JM. The next fails on HM and IL A D F H J L A E G I K M B D G I J L B D G I J M B E F H K L C D G H K M C D G H K L C E F I J M Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 25
Combination Testing • When all else fails, add rows. We need one for HM and one for IL, so add two rows. In general, we would need as many rows as the last column has values. • The other values in the two rows are arbitrary, leave them blank and fill them in as needed when you add new columns. At the very end, fill the remaining blank ones with arbitrary values • We have 8 tests instead of 3 x 2 x 2 x 2=96 Black Box Software Testing Copyright © 2005 A D F H J L A E G I K M H M B D G I J M B E F H K L I L C D G H K L C E F I J M Cem Kaner & James Bach 26
Let’s try this again on an old Netscape preference dialog Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 27
The Netscape example • If we just look at the Appearance tab of the Netscape Preferences dialog, we see the following variables: – Toolbars -- 3 choices (P, T, B) (pictures, text or both) – On Startup Launch --(browser, mail, news). Each is an independent binary. • Browser (Y, N) • Mail (Y, N) • News (Y, N) – Start With -- 3 choices (B, V, E) (blank page, valid existing file, error (syntax) in the URL) (Many more cases are possible) – Links -2 choices (D, U) (don’t underline, underlined) – Followed Links -- 2 choices (N, E) (never expire, expire after 30 days) (Many more cases are possible) Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 28
The Netscape example • I simplified the combinations by simplifying the choices for two fields. • In the Start With field, I used either a valid home page name or a blank. Some other tests for this field are: – Link to a different type of file, such as pdf – Link to a nonexistent file – Abbreviated URL, such as name. htm instead of http: // – File on the local drive, the local network drive, or the remote drive – maximum length file names, maximum length paths – Note that a bad URL won’t stop Netscape from starting, so we should be able to use an error case here without blocking testing of the other variables • For combination testing, select a few of these that look like they might interact with other variables. Test the rest independently. • Similarly for the Expire After field. This lets you enter the number of days to store links. If you use more than one value, use boundary cases, not all the numbers in the range. • In multi-variable testing, use partition analysis or other special values instead of testing all values in combination with all other variables’ all values. Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 29
All N-tuples • We can create 3 x 2 x 2 x 3 x 2 = 288 different test cases by testing these variables in combination. Here are some examples, from the combination table. • This is what Jorgensen would call “strong normal” testing. • Strong because we test for faults triggered by a combination of conditions. • Normal because we omit error cases. Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 30
Here are the 288 test cases. Every value of every variable is combined with every combination of the other variables. Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 31
All N-tuples When creating a combination table, I strongly recommend that you order the columns from the variable with the most values to the variable with the least. Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 32
All Singles • There are 3+3+2+2+2=16 different individual (single) values of interest. • We can cover them in 3 tests Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 33
What about pairs? • To simplify this, many testers would test variables in pairs, each test involving only 2 values. • There are 109 pairs in our example. • Testing only 2 variables at once is an inefficient form of combination testing. • One test that combines 7 variables incorporates 21 tests of pairs of variables. Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 34
Combinatorics “Combinatorics is, loosely, the science of counting. This is the area in mathematic in which we study families of sets (usually) finite with certain characteristic arrangements of their elements or subsets, and ask what combinations are possible, and how many there are. This includes numerous quite elementary topics, such as enumerating all possible permutations or combinations of a finite set. ” www. albany. edu/faculty/tangr/isp 602/notes/terms. htm • In combinatorial testing, we test many variables together as an efficient way of testing many of the combinations of those variables (e. g. testing 7 variables together in one test captures 7 C 2 = 21 tests of the pairs). • So how many tests would we have to run to cover all the pairs? Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 35
All pairs for Netscape We can cover all 109 pairs inside 9 tests Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 36
All pairs for Netscape • Let’s work it through. • We start with the first two variables (biggest and second biggest number of values of interest. ) • Here all the pairs of those two variables. There are 3 x 3 = 9 of them Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 37
All pairs for Netscape Add the next variable • We need all the pairs of the – 1 st and 2 nd variables – 1 st and 3 rd variables – 2 nd and 3 rd variables • We already have the pairs for 1 st & second variables • For the 1 st and 3 rd, we need – a Y with a P, an N with a P, – a Y with a T, an N with a T, – a Y with a B and an N with a B. • The values of the 3 rd variable for the other cases don’t matter for 1 st & 3 rd, but they might matter for 2 nd & 3 rd. Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 38
All pairs for Netscape Add the 4 th variable. We have the pairs for the 1 st 3, we just have to work in the 4 th. Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 39
All pairs for Netscape • Keep going, through the 7 th variable. Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 40
All pairs Reminder of a common misconception. The lower bound on the number of rows is the number of values of column 1 times column 2 but we often need more than that. Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 41
Combinatorial testing • At one of the LAWST meetings, we were advised that Microsoft often uses a modified all-singles in configuration testing: – All singles, plus – All other combinations designated by marketing (or by error history) as particularly interesting • Similarly if we use all pairs, we might add to the set of tests: – Special cases (marketing) – Special cases (identified risks of higher-order interactions) Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 42
Combinatorial testing • www. pairwise. org has been collecting references and links to tools, including free tools • Another free tool is at http: //www. satisfice. com/tools/pairs. zip • Rob Vanderwall has developed VPTAG, which allows you to specify some constraints (a given value of X makes a range of values of Y impossible). Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 43
Let’s add some complications • So far, we’ve assumed – Independent variables – All valid values are equivalent • What if we have multiple valid equivalence classes? • Let’s assume fixed precision, to 1 digit after decimal – Invalid: X < -100 boundary -100. 1 – Valid 1 -100 <= X < 0 bounds -100. 0, -0. 1 – Valid 2 0 <= X <=5 bounds 0. 0, 5. 0 – Valid 3 5 < X < 10 bounds 5. 1, 9. 9 – Invalid 10<= X bounds 10. 0 • These all become values of interest in combinatorial tests Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 44
Let’s add more complications • So far, we’ve assumed – Independent variables – All valid values are equivalent • What if the values of one variable affect the validity or effect of the values of another? Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 45
A common example: Testing a date field • 0 < day < it depends • 1 <= month <= 12 • 2000 < year < 3000 (whatever limits you choose) • For month 2 1 <= day <= 28 or 29 • For months 4, 6, 9, 11 1 <= day <= 30 • For months 1, 3, 5, 7, 8, 10, 12 1 <= day <= 31 • See Jorgensen for a thorough analysis Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 46
Let’s make this more challenging. The next slides present the Open Office Writer page style dialog. Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 47
These are interesting as a group because they all interact. Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 48
Just print a page. Its layout is jointly determined by all of these Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 49
So how do we test all of these? Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 50
Can you list the relevant variables? Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 51
How many variables are on this page? Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 52
The number of variables on this page depends on how many columns you choose. Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 53
At last, we’re through this one (1) dialog. You can see why people would give up and do all singles or random combination. Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 54
Thoughts on all pairs • All pairs is ideal for independent variables. – A classic use is configuration testing. – But if they’re independent, why test them in combination? • We are managing a type of coverage here. • We are rarely working from a theory of error. – Schroeder & Bach argue that in this case, we are probably as well off using a random combination algorithm. The set of tests will approximate all pairs – If we combination-test the program several times, randomness creates variation in the testing • All pairs is adaptable when the number of constraints is small – Whenever a test has an invalid pair, substitute two tests, identical except that you substitute a valid value for the first (second) value of the pair. All other pairs in the test stay intact and are tested Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 55
Thoughts on all pairs All pairs is inefficient when the number of constraints is significant. • Beizer (Black Box Testing) discusses the general case in which there are several levels for each variable and the program behaves differently as a joint function of the settings of several variables. His presentation of a domain testing approach to this problem is interesting but as described, I find it challenging to apply. • In electrical engineering, this situation is analyzed as Combination Circuit Testing. Given a set of values of interest for several variables (you arrive at them through domain analysis or in some other way), the question is whether the program behaves correctly for each combination. • In software testing, the analysis is called Cause-Effect Graphing Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 56
Alternative approaches There are several approaches to combination testing • Mechanical (or procedural). The tester uses a routine procedure to determine a good set of tests • Risk-based. The tester combines test values (the values of each variable) based on perceived risks associated with noteworthy combinations • Scenario-based. The tester combines test values on the basis of interesting stories created for the combinations. Black Box Software Testing Copyright © • Some groups of variables involve too complex a set of relationships for you to analyze (given your skills, tools and the time available) or are not well enough specified for you to analyze. • If you believe that you need to test combinations anyway, and want to consciously control the design of the tests, you might want a technique that helps you explore relationships and make sense of them. 2005 Cem Kaner & James Bach 57
Exploring relationships • Look at this record (bigger on the next slide), from the Timeslips Deluxe time and billing database. In this dialog box, click the arrow next to the Consultant field to edit the Consultant record (my name, billing info, etc. ) or enter a new one. • If I edit it here, will the changes carry over to every other display of this Consultant record? • Also, note that the End Date for this task is before the Start Date. That’s not possible. Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach
Exploring relationships Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach
Exploring relationships The program checks the End Date against the Start Date and rejects this pair as impossible because the task can’t end before it starts. The value of End Date is constrained by Start Date, because End Date can’t be earlier than Start Date. The value of Start Date constrains End Date, because End Date can’t be earlier than Start Date. Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach
Exploring relationships A relationship table Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 61
Relationship Table THE TABLE’S FIELDS • Field: Create a row for each field (Consultant, End Date, and Start Date are • • • examples of fields. ) Entry Source: What dialog boxes can you use to enter data into this field? Can you import data into this field? Can data be calculated into this field? List every way to fill the field -- every screen, etc. Display: List every dialog box, error message window, etc. , that can display the value of this field. When you re-enter a value into this field, will the new entry show up in each screen that displays the field? (Not always -- sometimes the program makes local copies of variables and fails to update them. ) Print: List all the reports that print the value of this field (and any other functions that print the value). Related to: List every variable that is related to this variable. (What if you enter a legal value into this variable, then change the value of a constraining variable to something that is incompatible with this variable’s value? ) Relationship: Identify the relationship to the related variable. Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach
Exploring relationships • Given the relationship, – Try to enter relationship-breaking values everywhere that you can enter V 1 and V 2. – Pay attention to unusual entry options, such as editing in a display field, import, revision using a different component or program • Once you achieve a mismatch between V 1 and V 2, – the program's data no longer obey rules the programmer expected would be obeyed, so anything that assumes the rules hold is vulnerable. – Do follow-up testing to discover serious side effects of the mismatch Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 63
Many relationships among data • Independence – Varying one has no effect on the permissible values of the other or on how the computer responds to a value of the other variable. • Causal determination – By changing the value of one, we determine the value of the other. For example, in selecting page layout, if you select “Letter” the page becomes 8. 5 x 11. • Constrained to a range – For example, width of a line must be less than the width of the page. – In a date field, the max day is determined by the month • Selection of rules – Example, hyphenation rules depend on the language you choose. • Relations are often reciprocal, so if V 2 constrains V 1, then V 1 might constrain V 2 (try to change V 2 after setting V 1) 64 Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach
Many relationships • Logical selection from a list – processes the value you entered and then figures out what value to use for the next variable. Example: timeouts in phone dialing: • 0 seconds on complete call 555 -1212 but 95551212? • 10 seconds on ambiguous completion 955 -5121 • 30 seconds on incomplete 555 -121 • Logical selection of a list: – For example, in printer setup, choose: • Office. Jet – get Graphics Quality, Paper Type, and Color Options • Laser. Jet 4 – get Economode, Resolution, and Half-toning. » Marick (Craft of Software Testing) discusses catalogs of tests for data relationships. Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 65
Complex Relationships Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 66
Data Relationship Table • Looking at the Word options, you see the real value of the data relationships table. Many of these options have a lot of repercussions (they impact many features). • You might analyze all of the details of all of the relationships later, but for now, it is challenging just to find out what all the relationships ARE. • The table guides exploration and will surface a lot of bugs. ------------------PROBLEM • Works great for this release. Next release, what is your support for more exploration? Black Box Software Testing Copyright © 2005 Cem Kaner & James Bach 67
Let’s sum up There are several approaches to combination testing • Mechanical (or procedural). The tester uses a routine procedure to determine a good set of tests • Risk-based. The tester combines test values (the values of each variable) based on perceived risks associated with noteworthy combinations • Scenario-based. The tester combines test values on the basis of interesting stories created for the combinations. Black Box Software Testing Copyright © Mechanical approaches: – Give you a handle on some complex problems – Provide easy justification for management. The number of tests needed is driven by theory and computed by the tool. Doesn’t appear discretionary. This is an important difference from random testing. – Provide an intuitively appealing coverage model – Appeal to the mathematically inclined – Are rarely based on a plausible theory of risk. (They’re wasteful, however, if and only if a risk-based model would generate substantially different or fewer tests. ) 2005 Cem Kaner & James Bach 68
- Slides: 68